[Mlir-commits] [mlir] [mlir] Speed up FuncToLLVM: CallOpLowering using SymbolTableCollection (PR #68082)

[llvmlistbot at llvm.org]

tdanyluk wrote:

Thanks for the quick comments you both.
The solution was indeed not perfect: it broke some tests.
Now my updated solution doesn't.

It turned out that we can't really use symbol tables at this point, because during the conversion there can be multiple operations with the same `SymName` (both `func::FuncOp` and `LLVM::LLVMFuncOp`).
So I opted to simply precalculate a map of function names to booleans.

This only speeds up `ConvertFuncToLLVMPass`, it leaves the existing logic for external callers of `populateFuncToLLVMConversionPatterns`.

BTW, sorry, this is my first MLIR commit, please LMK what should I improve.

https://github.com/llvm/llvm-project/pull/68082